T-N6OT-standalone-install-test-per-harness-claude-codex-cursor-node
State: planning/draft · Impact: high · Complexity: medium
The assembler (T-2A6R) proves the assembled tree installs and runs
standalone, but that’s one generic check. The first public release needs
one pass per harness the plugin actually ships to — Claude Code, Codex
and Cursor — plus a plain Node install with no harness at all, so a
harness-specific path-resolution regression (the class of bug
T-VAW3 fixes) is caught before the tag is cut, not after a user
reports it.
Approach
Section titled “Approach”- Add one standalone install-test target per harness under the assembler’s validation step: install the assembled pack into a clean Claude Code plugin cache, a clean Codex config dir, and a clean Cursor extension dir, each in a scratch directory with no checkout of this monorepo present.
- Add a fourth, harness-less target: a plain
npm i -g(or equivalent) Node-only install, exercising the CLI directly with no agent host at all. - In each target, run a smoke script that invokes the launcher (or the
per-skill
scripts/sdlcshim where the harness renders skills) and confirms it resolves the CLI and executes a trivial verb (sdlc --versionor equivalent) with a zero exit code. - Wire the four targets into the assembler’s standalone-validation step
(
T-2A6R) so a failing harness blocks the Phase 14 gate the same way a failing strict-check does. - Record each target’s pass/fail in the phase gate’s review report.
| Area | Note |
|---|---|
scripts/sdlc-mirror/assemble.ts | Where the four install-test targets are invoked |
solutions/ontological/cli/ | The launcher under test |
solutions/ontological/plugin/plugins/sdlc/skills/ | Per-skill shims under test on Claude/Codex/Cursor |
Acceptance criteria
Section titled “Acceptance criteria”- AC-1: a clean install of the assembled pack into a Claude Code plugin cache runs the CLI with no checkout of this monorepo present.
- AC-2: the same holds for a clean Codex config-dir install and a clean Cursor extension-dir install.
- AC-3: a plain Node-only install (no harness) runs the CLI directly.
- AC-4: all four targets run as part of the assembler’s standalone validation and block the Phase 14 gate on failure.
Out of scope
Section titled “Out of scope”- Building the harness-rendering machinery itself (agent-pants’ job) — this task only tests what agent-pants and the assembler already produce.
- Fixing path-resolution bugs the tests surface — that’s
T-VAW3’s job; this task is the test harness that catches regressions of the same class going forward.
Dependencies
Section titled “Dependencies”T-VAW3— the install tests are the ones that catch a regression of the class this task fixes; a harness-cache install can’t pass before it lands.T-2A6R— the assembler produces the tree these tests install; the targets hang off its standalone-validation step.
Depends on
Section titled “Depends on”T-VAW3, T-2A6R